Combination of magnetoencephalographic and clinical features to identify atypical self-limited epilepsy with centrotemporal spikes

•Atypical SeLECTS has characteristic changes in MEG at an early stage.•Atypical SeLEECTS has interaction anomalies in the triple-network model.•MEG combined with clinical features at baseline can identify atypical SeLECTS. Our aim was to use magnetoencephalography (MEG) and clinical features to early identify self-limited epilepsy with centrotemporal spikes (SeLECTS) patients who evolve into atypical SeLECTS (AS). The baseline clinical and MEG data of 28 AS and 33 typical SeLECTS (TS) patients were collected. Based on the triple-network model, MEG analysis included power spectral density representing spectral power and corrected amplitude envelope correlation representing functional connectivity (FC). Based on the clinical and MEG features of AS patients, the linear support vector machine (SVM) classifier was used to construct the prediction model. The spectral power transferred from the alpha band to the delta band in the bilateral posterior cingulate cortex, and the inactivation of the beta band in both the right anterior cingulate cortex and left middle frontal gyrus were distinctive features of the AS group. The FC network in the AS group was characterized by attenuated intrinsic FC within the salience network in the alpha band, as well as attenuated FC interactions between the salience network and both the default mode network and central executive network in the beta band. The prediction model that integrated MEG and clinical features had a high prediction efficiency, with an accuracy of 0.80 and an AUC of 0.84. The triple-network model of early AS patients has band-dependent MEG alterations. These MEG features combined with clinical features can efficiently predict AS at an early stage.